1. Evaluation of OMI total column ozone with four different algorithms SAO OE, NASA TOMS, KNMI OE/DOAS Juseon Bak 1, Jae H. Kim 1, Xiong Liu 2 1 Pusan National University 2 Harvard-Smithsonian Astrophysical Observatory (SAO) NASA GEOCAPE-Korea GEMS joint meeting March 21-24, 2013

2. Subject Optimal Estimation (OE) based ozone profile algorithm developed at SAO (Xiong Liu), which will be adapted/modified for GEMS algorithm. Evaluating the algorithm performance in retrieving total ozone column from OMI measurements through comparison with ground-based observation. Qualities of SOE total column ozone estimates are compared with three kinds of OMI operational products from TOMS v8.5, DOAS and OE algorithms algorithmPrincipalFitting windowDeveloperVersion SAO OEOE270-330SAO TOMS 317.5, 331.2NASA8.5 DOAS 331.1-336KNMI2.2.1 KNMI OEOE270-330KNMI1.1.1 In principle, OE-based profile algorithms should be able to provide more accurate total ozone estimates than the two total ozone algorithms because of its use of a wider wavelength range (270-330 nm) than those that are used for total ozone [Bhartia, 2002]. However, the successful performance of hyperspectral profile retrieval algorithms can be accomplished with accurate calibration and simulation of measurements and gook knowledge of measurement errors and the a priori covariance matrix [Liu et al., 2005; Liu et al., 2010a]

3. Reference : Brewer Measurements Brewer daily observation from WOUDC network - Period : 2005-2008, 36 stations over North Hemisphere Criteria for Brewer data -Only ‘DS (Direct Sun)’ observation mode Criteria for OMI data -Solar zenith angle < 80°, cloud fraction < 0.8, AI < 2 -Excluded pixels affected by row anomalies Criteria for collocation between OMI and Brewer -Within 150 km in the radius between center of OMI and ground station -Same day (Note: Brewer data is given as daily averaged product) -We didn’t average OMI pixels within 150 km, but take one closest pixel

4. 1. Station to station comparison OMI – Brewer (%) ● double + Single 36 sites were initially selected from WOUDC, providing at least 100 days with DS measurements every year 29 sites were finally selected where OMI and Brewer comparison shows no noticeble deviations from neighbouring stations through the similar selection procedure done in Balis et al. (2007)

5. 2. Solar Zenith Angle dependency - Negative correlation between bias and SZA at SZA 60° at SZA 60° - SZA dependency increase for clear sky condition and is unchanged for different OMI cross track position groups but Large offset exists. - Negative correlation between bias and SZA - Dependency (SZA > 40 °) slightly increase at high CF group - Positive correlation between bias and SZA - Dependency (SZA < 60 °) largely increase at high CF group and nadir pixels (small VZA) - No SZA dependency - Typically, bias within ± 0.5 % - Dependency is unchanged for any condition.

6. 3. OMI Cross Track dependency KNMI OE - Strong cross track position dependency of mean biases : 4 % at nadir and 1 % at extreme nadir-off - Apply no tuning for level1b KNMI DOAS - Large fluctuations at 4, 16, 20, 26 positions - Apply no tuning for level1b SAO OE -Apply a Soft calibration to OMI level 1b radiances in order to correct possible all kinds of calibration errors including those causing cross-track/wavelength dependent biases as well as part of straylight error. NASA TOMS - Apply a Soft correction to take out across variability Note : KOE and SOE is retrieved at UV1 spatial resolution and TOMS and DOAS at UV2. 1 and 2 for UV2 is defined as 1, 3 and 4 for UV2 is defined as 2.

7. The feature of cross track-position dependent biases shown in KOE TO - Brewer comparison is distinctly found in the SCO/TO difference between SOE and KOE. On the other hand, TCO differences typically increase from 1.5 % at nadir positions to 3 % or more toward nadir-off positions.  the existence of cross-track dependent calibration problems in both UV1 and UV2 channels which could affect the ozone retrievals in both stratosphere and troposphere. the SOE soft calibration helps to remove the large negative biases at the extreme left side of cross- track positions, but could not explain the large differences in the cross-track position dependence between SOE and KOE algorithms. Probably, other implementations such as using different ozone cross section data set, a priori covariance matrix, and radiative transfer model could be secondary cause to the different cross-track position dependence. 3. OMI Cross Track dependency

8. Direct comparison in fitting residuals between SOE and KOE Lat = 31.5, SZA = 30.4 Lat = 82.8, SZA = 74

9. Direct comparison in fitting residuals between SOE and KOE Note : retrievals of one orbit [15739, 2007m0701t01] are used for comparison (SZA < 85°) Glint effect SAOKNMI

10. Note : solar zenith angle < 45 。 considered Dependency on Cloud fraction DOAS significant negative dependent bias TOMS mild positive dependent bias SOE remarkable stability for both clear/cloud conditions with one exception at the bin of 0.95 – 1.0 one exception at the bin of 0.95 – 1.0 KOE Larger bias under partly cloud (0.2<CF<0.8) could be related to a switch point in the algorithm between could be related to a switch point in the algorithm between fitting the surface albedo and fitting the cloud albedo. fitting the surface albedo and fitting the cloud albedo. Commonly increasing scatter With increasing cloudiness Dependency on CTP A stable variability in total ozone relative differences for low and middle clouds for all algorithm. and middle clouds for all algorithm. Large deviations for high cloud ( CTP < ~ 350 hPa), but SOE produces the smallest deviations. SOE produces the smallest deviations. 4. Cloud parameter dependency

11. 5. Monthly relative difference SOE and DOAS show similar seasonal variability likely due to the common o3 Cross section data (BDM) : TOMS and KOE uses Bass and Paur (1984) and Maicet e al. (1995), respectively. Seasonal dependency is a quite small for mid latitudes, but significant at high lat. with Winter maxi. & Summer min. High correlation between the temporal changes between KOE biases and solar zenith angle : Strong negative correlation at mid latitude and strong positive correlation at high latitude, which is consistent with negative dependency on small SZAs and positive dependency on high SZAs. SOEDOASKOETOMS SOE 1 DOAS 0.881 KOE 0.16-0.041 TOMS 0.790.720.431 SZA 0.320.49-0.82-0.00 SOEDOASKOETOMS SOE 1 DOAS 0.711 KOE 0.230.101 TOMS 0.350.260.731 SZA 0.300.35-0.68-0.62 SOEDOASKOETOMS SOE 1 DOAS 0.831 KOE 0.570.261 TOMS 0.170.09-0.051 SZA0.800.690.55-0.09

12. Summary & Conclusion Excellent performance of SOE algorithm is demonstrated by showing a remarkable stability w.r.t OMI geometries and cloud properties, long-term/seasonal stability TOMS : mild negative and positive dependency on SZA and CF. DOAS : stronger positive and negative dependency on SZA and CF. Need a very hard understandings to GEMS radiometric/wavelength calibrations to apply the SAO algorithm.  KOE –Brewer shows large dependency on OMI geometries ( SZA, cross track position)  High correlation in monthly averages between KOE biases and SZA